COILS - BUGFIX - Use psilim on the external-forcing path too, not just the coil path

The coil branch of PerturbedEquilibrium already passed psi=ffs_intr.psilim
(PerturbedEquilibrium.jl:98), but the other branch — external forcing data read
from file — called convert_forcing_normalization! with no psi, so it fell back
to the default surface. That path therefore normalized on the wrong surface
silently, with no error, whenever dmlim/qhigh/psiedge moved psilim inward of
psihigh (e.g. 0.998902 vs 0.9995 on a DIII-D case: a 6e-4 gap, same order as the
psi_N=1 extrapolation this PR already fixed).

Both production paths now use psilim, matching Fortran (gpec/gpec.f:431
field_bs_psi(psilim,...)). psilim is the ODE's last grid point to within eps=1e-10
(the final chunk ends at psilim*(1-eps), EulerLagrange.jl:504) and is snapped to
psi_store[end] exactly in the dW-peak truncation path (Riccati.jl:1375,1870,
EulerLagrange.jl:230).

The psihigh default on the three signatures is retained and re-documented as the
bare-PlasmaEquilibrium fallback — reachable only by direct callers with no
ForceFreeStates solve, for whom psilim does not exist yet — not as the intended
control surface.

Full test suite passes locally.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Author: d-burg

src/ForcingTerms/CoilFourier.jl

@@ -81,13 +81,17 @@ extrapolated distance is a fixed 1 − psihigh while the final spline interval s
 packing, so the error grows with resolution — which is why this surfaced only once the pass-1
 auto grid was refined.
 
-NOTE: strictly faithful is `psilim`, the *integration* limit, not `psihigh`, the *equilibrium
-spline* limit. They are equal unless `dmlim`/`qhigh`/`psiedge` truncation fires, in which case
-`psilim < psihigh`. PPPL shipped a fix for exactly this confusion (`docs/releases.rst:281`:
-"Fixes inappropriate uses of psihigh, which may not be the end of integration psilim if
-sas_flag, qhigh, or peak_flag are used"). `psihigh` is used as the default here because it is
-all a bare `PlasmaEquilibrium` knows; callers holding a ForceFreeStates solve should pass
-`psi = intr.psilim` explicitly, as the benchmarks do.
+NOTE ON THE DEFAULT: the physically correct control surface is `psilim`, the *integration*
+limit, not `psihigh`, the *equilibrium spline* limit. They are equal unless
+`dmlim`/`qhigh`/`psiedge` truncation fires, in which case `psilim < psihigh`. PPPL shipped a fix
+for exactly this confusion (`docs/releases.rst:281`: "Fixes inappropriate uses of psihigh, which
+may not be the end of integration psilim if sas_flag, qhigh, or peak_flag are used").
+
+**Both production paths pass `psi = ffs_intr.psilim` explicitly** (PerturbedEquilibrium.jl, coil
+and forcing-file branches), as do both Fortran-comparison benchmarks. This default of `psihigh`
+therefore applies only to direct callers holding a bare `PlasmaEquilibrium` with no
+ForceFreeStates solve — for which it is the outermost surface that exists, and always a strict
+improvement on extrapolating to ψ_N = 1. Pass `psi` explicitly whenever `psilim` is known.
 
 The coupling surface therefore moves inward by (1 − psihigh): 0.05 % of flux at the default
 psihigh = 0.9995, 0.5 % on the example, which lowers psihigh to 0.995 to capture q=6.

src/PerturbedEquilibrium/PerturbedEquilibrium.jl

@@ -107,8 +107,12 @@ function compute_perturbed_equilibrium(
                 modes_n = filter(m -> m.n == n, intr.forcing_modes)
                 isempty(modes_n) && continue
                 m_vals = [m.m for m in modes_n]
+                # Same control surface as the coil branch above: psilim, the integration
+                # limit (Fortran: gpec/gpec.f:431 `field_bs_psi(psilim, ...)`). Without it
+                # the normalization was taken on the equilibrium-spline limit, which differs
+                # whenever dmlim/qhigh/psiedge truncation moves psilim inward.
                 convert_forcing_normalization!(modes_n, norm_tag, equil, n,
-                    minimum(m_vals), maximum(m_vals))
+                    minimum(m_vals), maximum(m_vals); psi=ffs_intr.psilim)
             end
         end
     end